Desk
NVIDIA hardware, software, and developer news.
Nvidia's pitch for RTX Spark is that serious AI work, including always-on agents, now fits in a three-pound laptop, and this week's model drops make that pitch more concrete.
NVIDIA's open transcription model does speech, speakers, translation, and questions in one checkpoint, and ships the data pipeline to adapt it, which is the part that matters most.
NVIDIA is turning GPU sharing from a polite agreement into an enforced limit, which could quietly make a lot of expensive hardware go further.
NVIDIA is betting that the next Windows gaming laptop runs on Arm, renders with neural networks, and talks back through small local models instead of the cloud.
Speaking French is not the same as knowing how a French passport office works, and NVIDIA's latest recipe targets that gap with synthetic residents and strict rules.
A dementia research consortium is pointing AI at the data that never gets published, and that may matter more than any bigger model.
NVIDIA's DGX Spark tuning guide shows big local-inference speedups are real, but the honest headline is 42.5% on real work, not 185% on a synthetic test.
General Compute is betting that the way to survive a shakeout among AI clouds is to stay small and flexible, run on non-Nvidia inference chips, and plug in wherever there's spare power.
Bengaluru’s Simplismart raised a $14 million Series B, with Nvidia participating, for software that tunes AI inference to get more from existing GPUs. Backing efficiency makes sense for Nvidia if cheaper inference creates more AI demand.
Meta is betting that open, Ethernet-based networking can wire together gigawatt-scale AI clusters, and it is giving away pieces of the stack to make that bet stick.
NVIDIA's Minerva points machine learning at the AI data center itself, chasing the quiet hardware faults that waste GPU hours, and it's a good use of AI that also deepens NVIDIA's lock on the stack.
NVIDIA's Mission Control 2.5 brings automatic job and hardware recovery to Kubernetes, a practical step toward self-healing AI clusters, with the restraint to not overreact being the smartest part.